7 April 2020
Correction
Mea culpa: Yesterday’s post contained a math error in, well, my math example. (Thanks to P for pointing it out!). It’s now updated with the correct figures. Importantly, in my made-up example about the risk of false positives from serological tests, 50% (not 8%) of all positives would be false positives.
COVID-19 in US children
From the Johns Hopkins Center for Health Security newsletter (emphasis in bold mine):
COVID-19 IN CHILDREN The US CDC COVID-19 Response Team published a study of COVID-19 disease presentation and severity in US children. The study, published in the CDC’s Morbidity and Mortality Weekly Report, analyzed clinical data for 149,082 COVID-19 patients in the United States reported between February 12-April 2, including 2,572 pediatric patients. Pediatric patients represented only 1.7% of those patients. Compared to adult COVID-19 patients, fewer pediatric patients experienced fever, cough, or shortness of breath (73% vs 93%). With respect to disease severity, only 5.7% of pediatric patients were hospitalized, compared to 10% for adult patients. Three deaths were reported among the pediatric patients, but investigation is still ongoing to determine if COVID-19 was the likely cause of death in these patients. The study also provides analysis of underlying conditions in pediatric patients. Among those with available data (345 cases), 23% had at least 1 underlying health condition, with chronic lung disease (including asthma), heart disease, and compromised immune system being the most commonly reported. These data support the current understanding that COVID-19 disease tends to be more severe in adults than in children; however, severe disease and death can still occur in pediatric patients.
Are death counts commensurable?
We already knew that it was misleading to compare the count of confirmed cases across countries, both because the standards and lags for reporting vary by country, but also because the degree and criteria of testing, and therefore the degree of underreporting, varies widely.
Increasingly, it seems like this could be the same for the death toll — which until now has been taken by many epidemiologists and modelers as a more reliable figure. (For example, several papers try to estimate the true number of cases per country by backing it out of the reported death toll and applying estimates of lag time and fatality rates.)
For example, the FT reports today that the death toll in England may be 76% higher than previously reported:
The daily death toll in England from coronavirus was almost 80 per cent higher than the hospital figures reported during the accelerating phase of its spread across the country. Even these figures, running up to March 27 and verified by the Office for National Statistics, are an underestimate as they do not capture the total number of those who died with Covid-19 symptoms outside hospitals.
For what it’s worth, when I checked the various UK sources today (7 April 2020, at 8:46pm UK time), as a case study, I found the data on deaths to be consistent where it should be consistent, and the distinctions between sources clearly called out.
The Department of Health and Social Care data (on www.gov.uk) stated that this was deaths that had occured in hospitals with a positive test for COVID-19. The ONS data clearly includes deaths in nursing homes, so naturally is higher.
There have similarly been reports that Italy is not counting a significant number of deaths outside of hospitals; e.g., here.
I am not certain, but think that “deaths in hospitals for confirmed COVID-19 cases” may be something like a standard for at least ECDC data. (Does anyone have a source that can confirm this?)
It’s also worth noting that when Our World in Data compared three of the primary data sources for confirmed cases and deaths (WHO, Johns Hopkins, ECDC), they found that over time the discrepancies were very small.
So it seems to me that what is happening here is easy to understand, but important to remind ourselves of frequently: in the important effort to have data that is as comparable across countries as possible, the data of necessity becomes less comprehensive: we need to work to something like a lowest-common-denominator data set.
Even considering these limitations, I do think that comparing deaths across countries is helpful, once the important caveats are in place. That said, in the future, as we start to look at developing countries alongside wealthier countries, these definitions will not give as useful a comparison.
On a separate but related point, we also need to remind ourselves that the way we are attributing deaths to COVID-19 today will both understate and overstate the death toll, and is inconsistent with how we think of a disease like seasonal flu.
The attribution methodology will overstate deaths due to COVID-19 because many people who die with COVID-19 would have died in any case, some within a relatively short period. Some studies have found high rates of comorbidity with COVID-19.
It will understate deaths due to COVID-19 in cases where we did not have a test, where the test was a false negative, if the death was not recorded at all, or if the death was excluded from official statistics (for example, because it took place outside of a hospital).
Longer-term, what we really care about with a disease like COVID-19 (or seasonal flu) is excess deaths: deaths that we should causally attribute to COVID-19, because we believe that in the absence of the disease the death would not have taken place. This, as I understand it, is how we talk about deaths from seasonal flu.
This Economist article, which I’ve linked to before, gives reasons to believe that the excess deaths could be significantly higher than the reported deaths. It tries to identify excess deaths by simply comparing actual trends to expected trends in impated areas.
There’s another sense in which the deaths causally attributable to COVID-19 could be understated. Some people will die of unrelated conditions that might otherwise have been treated successfully, but were not treated because of the strain on the health care system or because of the lockdown.
How is the effective reproduction number evolving?
I found this new resource, from the CMMID Covid working group, fascinating. It attempts to estimate how the effective reproduction rate, Rt, is changing over time by country. (See here for a definition of Rt vs R0.)
When looking at the charts, keep in mind that an Rt over 1 means exponential growth, and an Rt under 1 means the epidemic is declining. Yes, a lower Rt is always better than a higher one, but we need to get, and stay, under 1.
These charts say that the vast majority of countries are either likely or very likely to be above 1:
In general, the country-specific charts demonstrate real progress, but also that we still have a ways to go in some countries.
Subjective Probability 